Temperature prediction using a Neofuzzy neuron approach

Key-Words: - Neofuzzy Neuron, Temperature prediction, artificial neural networks, fuzzy logic, Artificial

Intelligence.

1 Introduction

Artificial intelligence [2] and its diverse techniques

have been widely used for creating diverse prediction

models [4] including virtual sensors [10], formal

prediction model for identification and control

systems [16], among others. Weather forecast is a

very important area for building prediction models,

because its related with security, environmental

behavior and its impacts in basic activities as

agronomy, engineering, tourism, constructions,

social development, among others [12].

Some interesting contributions in the area of

utilization of artificial intelligence for the prediction

of weather variables can be found in [6, 13, 14].

Neofuzzy neurons approach was presented by

Takeshi Yamakawa [17, 18] and tries to combine the

best characteristics and capabilities of Artificial

Neural Networks [8] and fuzzy logic [19]. Some

applications of this neofuzzy neurons have been done

in areas as Time series Forecasting [5], virtual

sensors design for oil production processes [7],

identification of nonlinear dynamic systems [9], fault

detection and isolation [11] and operational condition

prediction in mechanical systems [15].

In this paper it will be presented a proposal for

building a prediction model for environmental

temperature using the neofuzzy neuron approach and

making some changes to the original algorithm in

order to improve the convergence time and the

accuracy of the prediction models.

This paper is organized as follows: Section 2 contains

the Neofuzzy neuron description and characteristics.

In section 3 it will be presented the proposal for

Temperature prediction using the Neo fuzzy neuron-

based approach and its utilization for predicting the

temperature in Ibarra city in Ecuador. In section 4

1,3,4FRANCKLIN RIVAS-ECHEVERRÍA, 2EDMUNDO RECALDE, 3IVÁN BEDÓN,

3STALIN ARCINIEGAS, 3DAVID NARVÁEZ

1Laboratorio de Sistemas Inteligentes

Universidad de Los Andes

Mérida, Edo. Mérida

VENEZUELA

2Escuela de Ciencias Agrícolas y Ambientales

Pontificia Universidad Católica del Ecuador-sede Ibarra

Ibarra, Provincia Imbabura

ECUADOR

3Escuela de Ingeniería

Pontificia Universidad Católica del Ecuador-sede Ibarra

Ibarra, Provincia Imbabura

ECUADOR

4Programa Prometeo

Secretaría de Educación Superior, Ciencia, Tecnología e Innovación

Quito, Provincia Pichincha

ECUADOR

Abstract: - In this paper it’s presented a temperature prediction application using a modified neofuzzy

neuronbased approach. This approach is an easy and accurate method for obtaining prediction results

using climatic measurements from the previous days. The variables used for building the model are

Temperature, Humidity, Dew Point, Wind speed, Pressure, Rain and Solar Radiation. It’s also

presented the obtained results for temperature prediction in Ibarra, Ecuador using three years data.

EQUATIONS

DOI: 10.37394/232021.2022.2.2

Francklin Rivas-Echeverría,

Edmundo Recalde, Iván Bedón,

Stalin Arciniegas, David Narváez

E-ISSN: 2732-9976

Volume 2, 2022

will be presented the conclusions, recommendations

and further works.

2 Neofuzzy neuron description

Neofuzzy neuron [17, 18] is a very simple structure

that uses the capabilities of artificial neural networks

and fuzzy logic and is a great tool for modeling

complex systems because of the simplicity of its

structure, composed by a single neuron that in its

weights are defined some fuzzy partitions in order to

model the complexity and nonlinearities of the

systems, being only necessary to change the number

of fuzzy partitions in the input variables, allowing

this way to find the most suitable structure. While in

artificial neural networks it is necessary to change the

number of layers, the number of neurons in each layer

and the activation function to find the appropriate

structure for obtaining good fitness, in the neofuzzy

neuron there is only one structure and the only

parameter that have to be changed is the number of

fuzzy segments.

In Figure 1 it can be seen the structure of a Neofuzzy

Neuron, where the interconnecting weights (synapse)

are replaced by a set of nonlinear functions fi, and the

cellular body perform the sum of the synaptic signals.

Figure 1: Structure of a Neofuzzy Neuron

In Figure 2 it’s depicted the structure of each of the

nonlinear functions fi. These functions are composed

of IF <condition>-THEN <action> rules, using as

<condition> the membership’s function value of the

input signals that are included in each of the

complementary fuzzy segments defined in Figure 3.

The <action> is a singleton with wij as corresponding

value.

Figure 2. Structure of the Nonlinear Function (synapse)

Figure 3. Complementary fuzzy segments

For obtaining the corresponding fi(xi) (output value

of the synapse) it’s used a defuzzification process that

consider the complementary structure of the

segments (the sum of the two activated membership

functions should be equal to 1). So, the neofuzzy

neuron output may be given as follows:

1,1, )()()

(++

+= kiikiikiikii wxwxxf

µµ

(1)

Where:

)( iik x

is the membership value obtained for the

input signal xi.

w are the interconnecting weights.

The incremental updating (Stepwise Training)

learning algorithm used for updating the weights, is

as follows:

)()(

ikijkkij

xtyw

−

∆

(2)

where:

yis the neofuzzy neuron output.

t is the desired output.

fi(xi)

µi1

wi1

µij

wij

µin

win

min max

µij

EQUATIONS

DOI: 10.37394/232021.2022.2.2

Francklin Rivas-Echeverría,

Edmundo Recalde, Iván Bedón,

Stalin Arciniegas, David Narváez

E-ISSN: 2732-9976

Volume 2, 2022

is the learning rate.

In this particular work it has been used two different

learning rate values: It was used a bigger value for

the beginning of the algorithm in order to find a faster

convergence and then it was uses a smaller value in

order to have a better fitting and more accurate

predictions.

3 Temperature prediction using the

Neofuzzy neuron-based approach

It was used the neofuzzy neuron-based approach for

predicting the temperature in Ibarra city [20], that is

located in Provincia de Imbabura in Ecuador. Ibarra

is a very particular city because it has many climatic

changes during the day and is an interesting place for

constructing a temperature prediction model.

As it was presented in [12] it was first studied the

climatic station and the variables that have been

measured during more than seven year. After that, it

was studied the relationship between the variables

and it was selected next variables for creating the

temperature prediction model: Previous days

Temperature, Humidity, Dew Point, Wind speed,

Pressure, Rain and Solar Radiation.

For predicting the temperature for one day in the

future, it was used information concerning the

measurements from one day before and also with

information from previous days up to five previous

days. The general structure of the neofuzzy neurons-

based approach model can be seen in figure 4, where

x1, x2, x3, x4 and x5 correspond respectively to the day,

month, year, hour and minute to be predicted. x6 is

the previous day temperature, x7 the previous day

humidity, x8 the previous day Dew point, x9 the

previous day wind speed, x10 the previous day

pressure, x11 the previous day rain and x12 is the

previous day solar radiation. For a more general

model, where it’s desired to make the prediction

using more previous days, it will be used

5+7*previous days inputs, because the first 5 inputs

variables will correspond to the day, month, year,

hour and minute to be predicted and it will be

required 7 variables (Temperature, Humidity, Dew

Point, Wind speed, Pressure, Rain and Solar

Radiation) for each or the previous days used for the

prediction model.

Figure 4. General structure of the neofuzzy neuron-based

approach for temperature prediction

It was analyzed the data taken every five minutes

from January 1st 2012 until may 15th 2015 [12]. It was

made statistical analysis [3] concerning outliers, data

imputations [1] and data sets selection for training

and testing the model.

It was selected the data set that was going to be used

for creating the neo-fuzzy neuron-based model

(240.000 patterns) and the data set used for testing

the model (116.640 patterns).

It was build models using diverse learning initial and

final learning rate and diverse inputs from one

previous day until five previous days and the model

that gave the better results for training and testing

phases was the one that uses the information

concerning 4 previous days and the results can be

seen in figure

The model that gave better result was the one created

using nineteen (19) inputs variables, which means

that it was used the information from two previous

days from the moment wanted to be predicted. It was

used as initial learning rate

=0.0001 and final

learning rate (after 1000 iterations) of

=0.00001.

The results can be seen in figures 5 and 6. In both

cases (training and testing data sets) the error found

between the real data and the predicted data is

10.37% and 9.08% respectively.

EQUATIONS

DOI: 10.37394/232021.2022.2.2

Francklin Rivas-Echeverría,

Edmundo Recalde, Iván Bedón,

Stalin Arciniegas, David Narváez

E-ISSN: 2732-9976

Volume 2, 2022

Figure 5. Result for the training data set using

temperature prediction model

Figure 6. Result for the testing data set using temperature

prediction model

4 Conclusion

In this work it was presented a proposal for creating

a temperature prediction model using a neo fuzzy

neuron approach.

This model was created using some modifications to

the typical neofuzzy neurons approach, changing the

training rate, having one with bigger value for

obtaining a faster convergence and a smaller one after

some iterations in order to have a more accurate

values.

This temperature prediction model proposed was

used for modeling the temperature in the Ibarra city

in Ecuador and it was found good results with a

particular selected structure.

It will be continued this research, comparing these

models with other neuronal and intelligent or hybrid

systems models in order to try to improve the found

results.

Acknowledgment: Authors want to thanks the support

given to this project by the Secretaría de Educación

Superior, Ciencia, Tecnología e Innovación of

Ecuador and Prometeo Program.

References:

[1] Afifi, A. and Elashoff, R. Missing observations

in multivariate statistics III: Large sample

analysis of simple linear regression. Journal of

the American Statistical Association. 64, 1969,

pp. 337-358

[2] Aguilar, J., Rivas F. Introducción a las técnicas

de computación inteligente. Mérida, Venezuela:

Editorial Meritec. 2001.

[3] Anderson, T.W. An Introduction to Multivariate

Statistical Analysis. Wiley Series in Probability

and Statistics. Third Edition. John Wiley & Son.

United Sates. 2003.

[4] Bravo, C.E., Saputelli, L., Rivas, F. Pérez, A.G.,

Nickolaou, M., Zangl, G., De Guzmán, N.,

Mohaghegh S.D., Nunez, G. State of the Art of

Artificial Intelligence and Predictive Analytics in

the E&P Industry: A Technology Survey. SPE

Journal. Vol 19. Nro 4. 2014, pp. 547-563.

[5] Collantes, J. and Rivas, F. Time series

Forecasting using ARIMA, Neural Networks and

Neo Fuzzy Neurons. Proceedings of 3rd WSEAS

Internacional Conference on Neural Networks

Applications. Switzerland, 2002.

[6] Dombaycıa, Ö. A., Gölcü, M. Daily means

ambient temperature prediction using artificial

neural network method: A case study of Turkey.

Renewable Energy. Volume 34, Issue 4, April

2009, pp. 1158–1161.

[7] Garcia, R., Camargo, E., Rivas, F., Colina, E. and

Luzardo, M. Water cut virtual sensor design

using a Neo-fuzzy neuron and Statistical

techniques in oil production. Proceedings of 10th

WSEAS International Conference on Systems.

Greece 2006.

[8] Hagan, M., Demuth, H., and Beale, M. Neural

Network Design. Hagan Publishing. 2002.

[9] Lee, M., Lee, S.Y. and Park, C.H. A new neuro

fuzzy identification model of nonlinear dynamic

systems. International Journal of approximate

reasoning. Vol. 10. 1994. pp. 29-44.

[10] López, T, Pérez, A., Rivas, F. Data Analysis

Techniques for Neural Networks- based virtual

sensors. Proceedings of 8th WSEAS Int. Conf. On

Neural Networks. Vancouver, Canada. 2007.

[11] Novoa, D., Perez, A. and Rivas, F. Fault

Detection scheme using Neo-fuzzy Neurons.

Proceeding of IASTED International Conference

on Intelligent Systems and Control. USA 2000.

EQUATIONS

DOI: 10.37394/232021.2022.2.2

Francklin Rivas-Echeverría,

Edmundo Recalde, Iván Bedón,

Stalin Arciniegas, David Narváez

E-ISSN: 2732-9976

Volume 2, 2022

[12] Rivas, F., Recalde, E., Bedon, I., Arciniegas,

S., Narvaez, D. Environmental temperature

prediction using a Data analysis and Neural

Networks methodological approach.

Proceedings of 16th International Conference on

Neural Networks (NN '15). Rome, Italy. 2015.

[13] Ruano, A.E., Crispim, E.M., Conceição,

E.Z.E., Lúcio, M.M.J.R. Prediction of building's

temperature using neural networks models.

Energy and Buildings. Volume 38, Issue 6, June

2006, pp. 682–694.

[14] Shank, D. B., Hoogenboom and McClendon,

R. W. Dewpoint Temperature Prediction Using

Artificial Neural Networks. Journal of Applied

Meteorology and Climatology. Volume 47, Issue

6, June 2008, pp. 1757–1769.

[15] Soualhi, A., Clerc, G., Razik, H. and Rivas,

F. Long-Term Prediction of the Operating

Condition of Bearings by the Neo-Fuzzy Neural

Network. Proceeding of the IEEE International

Symposium on Diagnostics, Power Electronics

and Drives. Spain 2013.

[16] Todorov, Y., Tersiyska, M. and Petrov, M.

NEO-fuzzy State-Space Predictive Control.

IFAC Papers on line. Dec. 2015. pp. 99- 104.

[17] Yamakawa, T. A Neo fuzzy neuron and it

applications to system Identification and

prediction of chaotic behavior”. In

Computational Inteligence: Imitating Life. IEEE

press. 1994.

[18] Yamakawa, T. et al. “A Neo fuzzy neuron

and it applications to system Identification and

prediction of the system behavior”. Proceedings

of 2nd International Conference on Fuzzy Logic

& Neural Networks”. Japan. 1992.

[19] Zadeh, L., Fuzzy Sets, Information and

control, Vol.8, 1965, pp.338-353.

[20] Ibarra city.

https://en.wikipedia.org/wiki/Ibarra,_Ecuador

Creative Commons Attribution License 4.0

(Attribution 4.0 International, CC BY 4.0)

This article is published under the terms of the Creative

Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en_US

EQUATIONS

DOI: 10.37394/232021.2022.2.2

Francklin Rivas-Echeverría,

Edmundo Recalde, Iván Bedón,

Stalin Arciniegas, David Narváez

E-ISSN: 2732-9976

Volume 2, 2022